finding
active
finding:held-out-reconstruction-r2-rises-monotonically-with-dictionary-width-sparsity-and-block-dimension-for-all-three-featurizersHeld-out reconstruction R² rises monotonically with dictionary width, sparsity and block dimension for all three featurizers.
Reconstruction cannot rank featurizers.
Source paper
extracted_from(2026) · Thomas Fel · Matthew Kowal · Mozes Jacobs · Dron Hazra +21
Neighborhood — ranked by edge-count
Papers (1)
paper
Claims (1)
claim
- Support-term savings.
Related by similarity (8)
cosine ≥ 0.65 · no typed edgeEntities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.
- Rank-1 tasks (K=32, β=2) show empirical transition width matching N_crit(k)=k staircase almost perfectlyfinding0.764Sharp test of utility ordering.
- On DINOv3, classification and segmentation tolerate R²≈0.8, while depth needs R²≈0.9 to retain 95% accuracy.finding0.756Task-dependent distortion floors.
- Frontier ranking differs from other metrics.
- Pirate steering demo.
- Imagine instruction raises property name in J-lens while J-orthogonalized property probe stays at baseline.finding0.752J-space privileged.
- Characterizes internal structure of the six scoring dimensions
- Interpretive synthesis of DIM and cone intervention successes
- Validates that steering effects are persona-specific rather than generic perturbation artifacts